-
J. Alvarez. A distribution-free valid p-value for finite samples of bounded random variables. 2024. [arXiv]
-
J. Alvarez. Confident magnitude-based neural network pruning. 2024. [arXiv], [code]
Complementary material and animations
Magnitude-based pruning in action. By inducing sparsity we can visualize that the magnitude of the weights in the corners are zeroed faster, capturing the visual attention at the center of the images in the MNIST dataset.
Confident magnitude-based pruning image segmentation example. For large values of the sparsification ratio the predicted masks with the pruned model (fourth column) deviate substantially from the predicted masks with the dense neural network (third column).